A Simple and fast method to detect garbage dumping using pedestrian attribute

This paper presents a new method for the effective detection of real-life garbage dumping actions. Garbage dumping has become a serious problem with the continuous increase of city populations around the world. While monitoring surveillance camera footage is commonly used to catch garbage dumping, this approach is often unable to recognize real-life garbage dumping actions, resulting in many false alarms. Thus, We propose a quick and easy way to detect a variety of garbage dumping actions using pedestrian attribute information. The proposed method includes a pedestrian detector that detects pedestrians and their attribute, and behavior discriminator that identifies a garbage dumping action by comparing the pedestrian attribute pattern in each image frame with a dumping behavior pattern DB. The performance of the proposed method is confirmed by experiments using a garbage dumping action data set similar to real-life situations. When compared with previous studies, the proposed method can be faster as it is simple and only makes deep learning calls once. The proposed method is shown to be capable of identifying pedestrian behavior at a speed of 15 fps, indicative of potential application to real CCTV.